Stability-AI/generative-models · error · ValueError

unknown sampler {params.sampler}!

Error message

unknown sampler {params.sampler}!

What it means

get_sampler_config builds the sampler object by branching on SamplingParams.sampler (euler, ddim, dpm_solver2, etc.). When none of the branches match it falls through to this ValueError at the end of the function. It is called by text_to_image, image_to_image and refiner, so any inference entry point with a bad sampler name hits it.

Source

Thrown at sgm/inference/api.py:363

            verbose=True,
        )
    if params.sampler == Sampler.DPMPP2M:
        return DPMPP2MSampler(
            num_steps=params.steps,
            discretization_config=discretization_config,
            guider_config=guider_config,
            verbose=True,
        )
    if params.sampler == Sampler.LINEAR_MULTISTEP:
        return LinearMultistepSampler(
            num_steps=params.steps,
            discretization_config=discretization_config,
            guider_config=guider_config,
            order=params.order,
            verbose=True,
        )

    raise ValueError(f"unknown sampler {params.sampler}!")

View on GitHub (pinned to e8cd657656)

Solutions

1. Use one of the supported values of SamplingParams.sampler as defined in sgm/inference/api.py (e.g. "euler", "ddim", "dpm_solver2").
2. Check the branch conditions in get_sampler_config for exact accepted strings and match case/underscores.
3. If a new sampler is required, add a branch in get_sampler_config constructing its Sampler object.

Example fix

// before
params = SamplingParams(sampler="k_euler", discretization="ddpm")
// after
params = SamplingParams(sampler="euler", discretization="ddpm")

VALID_SAMPLERS = {"euler", "euler_ancestral", "dpmpp_2m_sde", "dpmpp_sde", "ddim", "uni_pc"}  # per sgm/inference/api.py
assert params.sampler in VALID_SAMPLERS, f"{params.sampler!r} not supported"

SamplerId = Literal["euler", "euler_ancestral", "dpmpp_2m_sde", "dpmpp_sde", "ddim", "uni_pc"]
def is_valid_sampler(x: str) -> bool:
    return x in get_args(SamplerId)

try:
    out = text_to_image(params=params)
except ValueError as e:
    if "unknown sampler" in str(e):
        params.sampler = "euler"  # safe default
        out = text_to_image(params=params)

When it happens

Trigger: Calling text_to_image(SamplingParams(sampler="dpm++2m")) or any sampler string not in the if/elif chain of get_sampler_config — e.g. unsupported names borrowed from other libraries ("k_euler", "uni_pc") or typos.

Common situations: Porting sampler names from A1111/ComfyUI/comfy workflows where sampler naming differs; typos like "dpm_solver_2" vs "dpm_solver2"; building sampler strings dynamically from config files with stale names.

Understand the failure class

Background: "Invalid value" and "allowed values are" config errors: what your library rejected and how to fix it — this error's family across 41 libraries.

Related errors

• unknown discretization {params.discretization}
• Model {model_id} not supported
• Sampler and loss function need to be set for training.
• unsupported dimensions: {dims}
• unknown merge strategy {self.merge_strategy}

AI-assisted analysis of Stability-AI/generative-models@e8cd657656 (2026-08-29). Data as JSON: /api/errors/5ce5edcef3b875a9. Report an issue: GitHub.